TOWARDS CHANGE VALIDATION IN DYNAMIC SYSTEM UPDATING FRAMEWORKS

Abstract

Dynamic Software Updating (DSU) provides mechanisms to update a program without stopping its execution. An indiscriminate update that does not consider the current state of the computation, potentially undermines the stability of the running application. Determining automatically a safe moment, the time that the updating process could be started, is still an open crux that usually neglected from the existing DSU systems. The program developer is the best one who knows the program semantics and the logical relations between two successive versions as well as the constraints which should be respected in order to proceed with the update. Therefore, a set of meta-data has been introduced that could be exploited to explain the constraints of the update. These constraints should be considered at the dynamic update time. Thus, a runtime validator has been designed and implemented to verify these constraints before starting the update process. The validator is independent of existing DSU systems and can be plugged into DSUs as a pre-update component. An architecture for validation has been proposed that includes the DSU, the running program, the validator, and their communications. Along with the ability to describe the restrictions by using meta-data, a method has been presented to extract some constraints automatically. The gradual transition from the old version to the new version requires that the running application frequently switches between executing old and new code for a transient period. Although this swinging execution phenomenon is inevitable, its beginning can be selected. Considering this issue, an automatic method has been proposed to determine which part of the code is unsafe to participate in the swinging execution. The method has been implemented as a static analyzer which can annotate the unsafe part of the code as constraints. This approach is demonstrated in the evolution of the various versions of three different long-running software systems and compared to other approaches. Although the approach has been evaluated by evolving various programs, the impact of different changes in the dynamic update is not entirely clear. In addition, the study of the effect of these changes can identify code smells on the program, regarding the dynamic update issue. For the first time, the code smells have been introduced that may cause a run-time or syntax error on the dynamic update process. A set of candidate error-prone patterns has been developed based on programming language features and possible changes for each item. This set of 75 patterns is inspected by three distinct DSUs to identify problematic cases as code smells. Additionally, error- prone patterns set can be exploited as a reference set by other DSUs to measure own flexibility.